Zac Wellmer comments

Results 8 comments of


                                            Zac Wellmer

annealing bias

I've got a repo where I implemented the importance sampling for [ddpg PER](https://github.com/zacwellmer/ddpg/blob/PER/ddpg.py). I'm still unsure if it works. But lines 68 - 91 might be useful for you. It's...

Reward loss targets don't account for episodes that finish within n steps

https://github.com/oxwhirl/treeqn/pull/4 Here's what I think is a quick fix. There could probably be a quicker vectorized implementation though

Reward loss targets don't account for episodes that finish within n steps

@Greg-Farquhar I think the [np -> torch commit](https://github.com/oxwhirl/treeqn/commit/5bc920c9f89f40111d7ae29193e68ca43d9e2bbe) introduced a new bug. In [make_seq_mask](https://github.com/oxwhirl/treeqn/blob/master/treeqn/utils/treeqn_utils.py) the mask variable is being updated in place and will cause our tmp_masks variable to change...

Reward loss targets don't account for episodes that finish within n steps

looks good!

Non continuous multi-processing

Looks like if you path up the jiffer buffer then the finklesnout should allow non continuous block chain cloud computed multi-processing

Feature request: Time picker

Here's a simple one that re-uses shadcn's calendar component. I stole the time field from @uncvrd. https://gist.github.com/zacwellmer/316e38705b6534907c142af8ca21fa9d https://github.com/shadcn-ui/ui/assets/9603276/5d74a539-96a0-498c-ac41-7157133913b5

render_mode=False is useless. Fake option not to render

@whoIsTheGingerBreadMan did you ever fix this?

render_mode=False is useless. Fake option not to render

True and False only determines the render format (rgb_array vs human). Regardless of your choice the environment will still render. Is there a reason why it's necessary to render the...